There's a lot about this question to be found in books by and about Luca Turin (The Secret of Scent and The Emperor of Scent, respectively).
One of the bases of Turin's critique of theories that tie scent to molecular shape is that these theories have very poor predictive power, so that the design of new fragrance molecules requires extensive trial and error. Turin claims, furthermore, that his theory -- which ties scent to the vibrational frequencies of a molecule's constituent atoms -- has better predictive power. (See the front page of http://www.flexitral.com/ for an instance of this claim.)
I don't know how accurate these claims are about the relatively predictive success of the two theories, and I don't know if it's appropriate to use the predictive success of either theory as a way of deciding in favor of one theory or the other.
I guess the roundabout answer to your question is, they are trying to match chemicals to known structures, but this matching process, as it is usually performed, seems to require a degree of trial and error.